118: Born to Compete — 郭人杰, From a 1997-Born Robot-Vacuum President to Building General-Purpose Home Robots
Summary
郭人杰’s investable edge is not the “gifted-class” label, but the high-frequency experimentation and organizational leverage he proved at Dreame: in under four years, he and his team grew China domestic-brand sales from roughly RMB100M to RMB6B. China alone went from roughly RMB300M to RMB3B in 2022; he was demoted twice, first after stockouts and later after missing a profitability deadline, then earned his way back through new products and profit delivery. “The floor for the next iteration is the ceiling of the last” is both his growth playbook and the source of a high-attrition, execution-heavy culture.
Lexiang is not pursuing a scientist-style “0 to 1” breakthrough, but a consumer-robotics opportunity with early PMF that can scale from “0.5 to 10.” 郭人杰 defines his strengths as finding new channels, engineering products quickly, and taking them from one to ten, so the opportunity must support a global business and combine with AI and robotics without forcing a head-on fight against established giants with superior resources. “Make money to fund the dream” is the constraint: first keep the company alive with products offering relatively high certainty, then explore new home terminals.
The first developer robot has already given Lexiang a relatively concrete survival model: at roughly $1,000, the product has 30,000 lead-capture emails and 2,000–3,000 community users; selling 2,000–3,000 units a month could cover the company’s operating costs. Targeting students and geeks, it offers a low-cost sim-to-real platform for reinforcement learning and was planned for launch in the third quarter of the year referenced on the show. It was an unexpected demand discovery, not the endgame, but 郭人杰’s judgment was direct: “When we saw the opportunity, we didn’t hesitate.”
The bigger bet is a “mobile hardware terminal” for the home, but the first use case is not folding clothes or doing chores; it is making information exchange among elderly people, children, and pets more natural than using a phone. The robot needs to follow users, move quietly, and maintain the right viewing angle, with AI filling information gaps among family members; once the platform matures, companionship, education, and pet interaction could grow like apps. Lexiang might run its first exploration in November–December of the year referenced on the show, at a scale that could be “50 units, 500 units, or 500,000 units”; the concrete demand for a home terminal remains unproven.
A tracked robot closer to the consumer market has signed a WALL-E IP deal and is planned for September–October, with indoor/outdoor mobility, outdoor security, and child-following as its core tests. 郭人杰 believes tracks broaden terrain adaptability, a humanoid IP can improve first-glance purchase appeal, and mobile cameras could sit between fixed surveillance and a guard dog; “taking it outside to show people” is itself one possible purchase motive. These remain hypotheses to validate, not a proven mass market.
Lexiang’s potential advantage is consumer-hardware engineering and rapid testing; its obvious gaps are motion control, robot mechanics, and emotional models. At the time, the company had roughly 39 people, 32–33 of them in R&D, with about 12–13 coming from the robot-vacuum industry; its CTO spent years at Ecovacs, while reinforcement learning was developed with Silicon Valley team K-Scale Labs. 郭人杰 acknowledges that “if you can’t do it, you can’t do it,” and expects both sides to fill their software and hardware gaps over time, but partnership dependence, parallel product development, and young-team execution remain risks.
郭人杰 welcomes teams such as Vita Dynamics, Light Robotics, and Miaodong, as well as large robotics companies, helping the category converge, because his strengths become easier to deploy once competition is clearly defined. For investors, the key question is not who first declares itself “the Apple of the robot era,” but whether Lexiang can control multi-product R&D spending and turn first-generation market feedback into iteration speed. His success standard for the year was deliberately restrained: “Define a generation of product that we ourselves are satisfied with,” with unit sales not yet the primary outcome.
Deep dive
1. The gifted class gave 郭人杰 freedom to define his own problems
郭人杰 entered Xi’an Jiaotong University’s gifted class at 15 and faced almost no exam pressure. He recalls a classmate next door doing chemistry experiments in the dorm bathroom: “One weekend we woke up and found the bathroom on fire.” The freedom was far greater than in ordinary classes.
Looking back in his junior year, he realized that what truly engaged him was not any single academic discipline, but “making an organization better in my own way.” That became the original instinct behind his later management style and decision to start a company.
Moving across engineering, gold-alloy research, and financial economics did not push him toward pure finance. Even after graduating from the London School of Economics, he concluded that he had “no real feel for trading numbers” and wanted to build a business around a tangible product.
2. Two university projects already contained his entrepreneurial DNA
As captain of the university debate team, he thought it was “not very fair” that competition access depended on whose family could afford the travel. He linked alumni across roughly 20 years and built a fund with donations of RMB30,000, RMB50,000, and other amounts per person to send team members to competitions in Australia and Singapore.
Seeing the information gap between students in Xi’an and those on the coast, he led 9 people in running a WeChat account and lecture series. A referral model of “one generation makes 3, three make 9, nine make 27” eventually expanded to 7 Xi’an universities and may have helped “thousands” of students; only later did he realize that a nonprofit organization with no revenue could also be a startup.
3. P&G was designed as an accelerated course in running a business
郭人杰 decided early that he would start a company, but first broke “running a business” into two unknowns: how to build a product and how to sell it. After asking several alumni for advice, he chose P&G’s management-trainee program because it emphasized a Marketing-driven business.
He worked in Marketing, Sales, and as a product manager. To understand R&D, IT, supply chain, HR, legal, and other functions, he repeatedly scheduled conversations: “The first time they may not tell you anything. After three conversations they start talking. After eight, they tell you everything.”
After 7 months, he used more than 10 sheets of A3 paper to map how a business actually happens and resigned for the first time. From the start, the goal was learning, not climbing a corporate ladder.
4. P&G’s nearly 200-year survival taught him to study the floor
Around 2020, as new consumer brands attacked every category, 郭人杰’s question was not why the newcomers were strong, but how P&G, founded in 1837, had “managed to stay alive this long.” It could not have been the first time it faced a regional assault.
He spent another roughly 3.5 months studying SOPs, databases, and organizational structures, trying to understand how a company writes failure into its systems and prevents the same mistakes from recurring. His conclusion was that P&G turned competition into “a game of seeing who can live longer.”
His summary was that “system capability is greater than any single consumer company.” With early warnings, risk assessment, and operating systems, a company can wait for excess returns to compress into an efficiency contest before entering: a willingness to “be the last into the market.”
5. Cultural consistency raises consensus efficiency—and can amplify bad decisions
P&G’s long-standing preference for internal development over external hiring produced employees with highly similar profiles, values, and language. 郭人杰 believes this improves not only execution efficiency, but also the efficiency with which consensus forms from the top down.
程曼祺’s follow-up focused on the cost: consistency means less internal collision, so a wrong call at the top can cost the company its window. After Douyin, Kuaishou, community group-buying, and other channels rose suddenly, 郭人杰 observed P&G bringing in senior external talent because outside experience had become more valuable than preserving cultural purity.
6. Unable to pick a startup, he learned who was better at picking
After leaving P&G, 郭人杰 interviewed with 5 or 6 early-stage consumer or vertical e-commerce companies and found that “every one of them was good.” He viewed joining a startup as investing in himself, so an inability to distinguish among them meant he would probably choose wrong.
His answer was not to keep guessing, but to interview for an investment role at IDG and accept the offer, hoping to look for good companies from the perspective of a leading VC. IDG recognized Dreame, and a deep conversation with 俞总 eventually moved Dreame ahead of the other candidates.
This also explains Lexiang’s financing relationships: IDG knew him before he joined Dreame and participated in the angel round roughly 5 years later. Matrix, ZhenFund, Monolith, Oasis, and other institutions invested before the product and data existed, effectively betting on the person and his track record.
7. 俞浩’s “12x formula” matched 郭人杰’s work philosophy
At 23, 郭人杰 joined Dreame to run the China marketing team and immediately asked 俞浩 why he would trust someone who had worked for only 1 year and had never managed anyone.
As 郭人杰 recounts it, 俞浩 described speed as multiplication: work 3x as long as others, iterate 2x as often, and be 2x as smart, for a total of 12x. That was also his explanation for how Dreame’s high-speed motors could cover in 2 or 2.5 years a path that took others roughly 30 years.
郭人杰 already believed that, outside the most cutting-edge R&D, most work depends on the willingness to iterate, iteration speed, and the ability to capture new opportunities. 俞浩 simply stated it more precisely: “Isn’t that exactly how I see the world?”
8. “I’ve never had a real job” became an experimental system at Dreame
When 郭人杰 worried about his lack of management experience, 俞浩 replied: “I’ve never had a real job.” His broader logic was that without experience, you have no default rules and do not care what a company is “supposed” to look like.
If there were 10 possible paths, Dreame would not use experience to bet on one first. It would spend the minimum resources to test all 10, then concentrate resources on the path that worked and “make it number one, make it to the extreme.”
9. Dreame broke its world-number-one ambition into 3 single-point wins
俞浩 divided a manufacturing company into R&D, supply chain, and brand. R&D had historically been concentrated in the US, supply chains in China, and brand strength in Europe; he believed China’s engineer dividend would pull more R&D back to China.
On supply chain, Dreame shifted its center of gravity to Suzhou because, as 郭人杰 recounts, the city produced roughly 70% of the world’s vacuum cleaners. If R&D and supply chain could lead, the remaining question was whether the brand could become number one.
Dreame did not try to calculate the optimal coupling of every link in advance. It believed that “if every single point is number one, the sum cannot be number two.” 郭人杰 admits this can create internal friction and redundancy, but the target is exceptionally clear.
10. 郭人杰 sees a brand as longing for a way of life
His original view was that when consumers see Western coffee, water, or brands as more premium, the underlying difference is not the object itself but a longing for the other side’s lifestyle. After living in the UK, he began to believe that the direction would reverse.
When Dreame posted the same technology content on WeChat Channels, users often accused it of “copying” and said it was impossible. On Bilibili, the response was “Chinese technology is unbeatable—triple-tap,” showing the confidence gap between 2 generations of consumers toward domestic technology.
He reduces the path to a leading brand to 3 steps: make the product good enough that “you don’t disappoint consumers”; build reputation by doing the right things consistently across generations; then secure distribution and scale with rising consumer confidence.
11. Returning from emotional marketing to technology was a prerequisite for premiumization
Dreame once followed the industry into content about independent women and emotional promotional films. The results were not bad, but 郭人杰 says everyone from the founder to the employees felt awkward because “this doesn’t feel entirely like Dreame.”
Only after the company moved beyond pure survival did the team confirm its shared belief: technology, supply chain, and brand all had to be number one. “People who build airplanes going to sweep floors,” coined by the PR head at the time, became a technology narrative rather than 郭人杰’s personal improvisation.
程曼祺 asked whether “being yourself” conflicted with testing all 10 paths. 郭人杰’s answer was that the stages are different: survival prioritizes efficiency; scaling and premiumization require consumers to understand clearly “who you are.”
12. Dreame’s first breakout Douyin users were urban older consumers, not affluent mothers
Industry wisdom held that the ideal floor-washer customer was the affluent, image-conscious mother. Dreame targeted that group and saw poor efficiency, so the team discarded its priors and retested every segment. It unexpectedly found an explosion of demand among urban middle-aged and older users.
In hindsight, the explanation was straightforward: affluent mothers had heard of Tineco and saw Dreame as close to a white-label brand; older users might have heard of neither, so the brand gap was smaller, while they still had purchasing power and a willingness to try shopping on Douyin. 郭人杰 captured that demand first, then went back after the mainstream customer.
13. High incentives and high attrition solved the talent problem when brand pull was weak
Dreame was little known at the time, and Suzhou was not a marketing talent hub. The team recruited heavily from nearby third-tier universities and junior colleges, then screened rapidly on results rather than waiting for ideal résumés.
High incentives meant sharing business proceeds directly with operators. Top livestreamers could earn roughly RMB50,000 a month and would voluntarily keep streaming until 2 or 3 a.m. when sales were strong because “everyone was making money.”
The other side was that 60%–70% of people were typically eliminated after roughly 2 weeks; of 10 people doing the same job, only the top 2 stayed. 郭人杰’s boundary was: “If you do the work well, we’re brothers and comrades-in-arms. If you don’t, we’re still brothers, but you have to leave.”
He once said that “going to Tsinghua or going to a junior college is just a question of which year you started playing video games.” He was not denying the probabilities associated with education, but emphasizing that outside elite R&D, willingness, iteration, and results can reorder the hierarchy.
14. A multi-platform marketing infrastructure and Douyin’s tailwind delivered the first step-up
In the first 618 after 郭人杰 joined, the team shifted Douyin, Bilibili, Weibo, Zhihu, Xiaohongshu, and other platforms from scattered placements to systematic operations. Business grew roughly 7–8x year on year, earning him his first promotion.
Starting in July 2021, they saw Douyin e-commerce moving simultaneously from nonstandard goods into standardized categories such as appliances, and from white-label products toward brands. While category leaders worried about uncertainty, Dreame—ranked around 15th—was willing to go all in.
This matched his view of platform arbitrage: traditional search e-commerce requires years of accumulation, while a new platform can suddenly give a weaker player a chance to be reordered.
15. “Copy first, iterate later” depended on testing throughput, not creativity
郭人杰 admits that early Douyin content and formats were “all copied.” When the team did not know how to create, imitation had much higher initial efficiency than originality. The difference was whether it could test the next generation faster after copying.
One test cross-validated the short-video selling point, livestream script and demonstration, and target audience. Products or selling points changed 6 or 7 times a day on average.
His A/B testing rule was to extract 2 lessons from the best-performing creative and extend them into 10 versions. As long as “the floor of the next iteration is the ceiling of the last,” iteration itself becomes an advantage.
16. The stockout after a volume spike turned a first promotion into a first demotion
Dreame was doing roughly RMB8M–RMB10M in monthly sales, but at Douyin’s peak it could sell RMB1M in a single day. Even inventory equivalent to 2.5 months of sales, worth roughly RMB25M, supported only about 25 days.
The stockout lasted 5 or 6 months and exposed 郭人杰’s complete unfamiliarity with hardware rules such as 3-month material preparation, 6-month long lead times, and tooling. The company inserted an executive with years of Ecovacs experience between him and his boss, effectively lowering him by one level.
郭人杰 did not frame it as a power struggle. He said he felt guilty for creating risk for the company. The failure made supply-chain resilience a hard constraint on every growth plan that followed.
17. After being reinstated, the front end finally gained product-definition rights
In February of the following year, 俞浩 asked 郭人杰 before a livestream with Su Bingtian: “Have you learned yet?” Dreame had just completed roughly RMB2B in financing in September 2021, and recovery after the stockout carried much greater delivery pressure.
The new structure put 郭人杰 in horizontal control of China product lines including robot vacuums, floor washers, and cordless vacuums, while the previous head managed functions vertically. It was the prototype for the later BU structure.
More important was the authorization: “You define what product to sell.” Dreame shifted from a B2B perspective in which customers such as Xiaomi defined the product to one in which marketing, sales, and R&D started from end-user needs.
18. Dreame used multiple models to compensate for immature PRD judgment
Both the front end and R&D lacked mature experience with user products. 郭人杰 believed that if they followed a large-company process and wrote a complete PRD based on subjective judgment, it would “most likely be wrong.” The solution was to push multiple series and models directly into the market.
Outsiders criticized the company for having too many models and leaving consumers unsure what to choose. He accepted that the problem existed, but regarded it as a testing cost: use flexible channels to find the optimal product first, then concentrate resources behind it.
19. H12 turned the floor-washer business around with “full features, add drying, price 15% lower”
The team gave its first strategically important floor washer 3 requirements: cover every feature offered by competitors, add one marketable point of differentiation, and keep the price roughly 15% below competitors.
The final differentiator was drying, using technology already developed for robot vacuums. Marketing tied the machine’s feature set to the Dreame brand so consumers would remember more than a single cleaning benefit.
H12 took roughly 2.5 months to develop, launched in March, and became the category’s top-selling single product in June. 郭人杰 believes front-end testing is important, but harder to replicate is the back end’s ability to deliver hardware of acceptable quality at that speed.
20. Dreame left Douyin after 3 weeks to fight on the search-commerce battlefield
Douyin produced business whenever the company bought traffic, and traffic disappeared when it stopped. The team worried that the brand would disappear with the subsidy once the platform’s tailwind faded. Search weight, reviews, and fulfillment records on Tmall and JD, by contrast, could leave behind organic traffic.
Dreame changed short videos that previously routed users into livestreams to direct them to Tmall and JD instead. Despite platform pressure, it reached the number-one single product on Tmall and number two on JD in roughly 3 weeks, with about 15% share for the product.
郭人杰 defined the move as the transition from a “new consumer brand” to an ordinary consumer brand: not merely harvesting a new channel, but leaving cumulative assets in mainstream channels.
21. 俞浩 pushed robot vacuums early and built the second growth curve
When floor washers had just proven out, 郭人杰 wanted to make another floor washer to reinforce the core business. 俞浩 instead demanded that Dreame immediately build its own-brand robot-vacuum business, even though the former was roughly 50x larger at the time.
郭人杰 “didn’t understand” then, but later saw it as a strategic decision. Floor washers and robot vacuums eventually both reached number two in their categories; two number twos together made Dreame number one in intelligent cleaning overall.
This was another expression of “full encirclement”: leave no obvious gap in technology, features, channels, or categories. Individual points might be redundant, but the whole business would not depend on a single source of arbitrage.
22. Bilibili’s long-form format matched the robot vacuum’s complex demonstration
A robot vacuum that first cleans along the edges and then moves in a bow pattern takes too long to demonstrate for Douyin’s short-decision format. Bilibili users, however, were willing to watch a 15-minute teardown and understand the product in full.
The team had a creator draw the Riverside Scene at Qingming Festival on the floor with soy sauce, let it dry for 2 nights, and then had the robot vacuum clean it. The video remained on Bilibili’s overall ranking for roughly 2 weeks and reached number one at one point, selling about 7,000–8,000 units.
At an average ticket of RMB3,599, 郭人杰 says the video generated roughly RMB30M in GMV, making it the most effective piece of content in Dreame’s history.
23. Community group-buying and rolling presales converted a spike into deliverable growth
During the pandemic, 郭人杰 noticed that the best white-label merchants at capturing traffic were all doing community group-buying. He brought in 20 sophomore students to enter neighborhood groups, paying RMB5 for each group contacted and an additional reward for successfully posting a link, with roughly RMB100 commission per unit sold.
In one 618 campaign, they reached roughly 6,000 groups; 1,000–2,000 opened orders and sold around 5,000 units. Combined with more than 10,000 units from Bilibili, robot-vacuum volume reached roughly 20,000 units, lifting share from 0.x%–1% to about 5%.
To avoid another prolonged stockout, supply chain ran 15-day rolling presales: before one batch ended, the delivery window for the next had already opened. Spot inventory did not return until 618 was over.
24. The 2022 618 campaign moved Dreame from 15th to 5th in the category
郭人杰’s figures were: roughly 15% share in floor washers, roughly 5% in robot vacuums, and a China ranking that jumped from 15th to 5th. Growth during that 618 was about 10x, and monthly growth in the second half of the year “never fell below 10x.”
This was not a low-pressure victory. The team went nearly 3 months without sleeping before 2 or 3 a.m. Combined with the trauma of the first stockout, it did not dare celebrate until both categories had inventory. Then “everyone drank until they were unconscious.”
25. After the price war removed weak players, Dreame lost its value-for-money space
Dreame and Tineco cut prices aggressively during 618. International brands were largely cleared out because their products lagged while prices were higher, and smaller brands could not match the marketing spend. For a time, the market was left with 2 major brands.
Midea, Haier, Deerma, Joyoung, Supor, and other established companies then entered and drove prices down to roughly half. 郭人杰 realized that the value-for-money proposition that had helped Dreame enter the market did not hold against companies with much stronger resources.
郭人杰 still uses 俞浩’s summary: “Value for money cannot be replicated.” A startup must first find a price gap that large companies have not covered, but loses that advantage once the giants enter.
26. The substance of premiumization was escaping the need to stay cheapest forever
郭人杰 distinguishes “expensive” from “premium.” Dreame cannot raise prices through storytelling alone like a luxury brand; it must first prove that stronger functionality justifies a higher price, then gradually build technical recognition.
He believes fragmented media lowers the cost of repositioning a brand. If a company spends 2 consecutive weeks talking only about “Dreame premium floor washers,” consumers may not even remember its value-for-money positioning from 3 months earlier, so a new brand may not be necessary.
The team chose a “king of machines” version of premium value: a price roughly 10% higher, but several additional features that make the perceived incremental value greater than the premium.
27. The M series captured the high-price tier with a detachable vacuum function
Tineco was selling for RMB5,299, so Dreame launched the M series. The M stood for Mix: the motor could be detached and used as a vacuum cleaner, with most other functions unchanged.
Marketing did not avoid the extra RMB299. It asked instead, “Can RMB299 buy you a good vacuum cleaner?”—translating a higher price into a proposition that still felt economical. 郭人杰 says the product took the number-one position in the high-price segment.
Only after completing that price jump did the team, from late 2022 to early 2023, use “people who build airplanes going to sweep floors” to reinforce its technology identity and attempt to move from functional premiumization toward true premium status.
28. Multi-SKU channel coverage amplified scale and created price-chaos risk
H12 Pro added features and cost roughly RMB300 more, selling for RMB3,299 through Tmall, JD, and Douyin direct sales. H12 went to formal distributors. The stripped-down H12 Lite, at roughly RMB1,999, served communities, micro-merchants, and influencer channels.
程曼祺 worried that a lower-priced product under the same brand would damage premium perception. 郭人杰 cited the iPhone, where lower prices for older generations do not prevent newer generations from being premium, and argued that brand strength depends on continually launching innovative products rather than keeping every generation at the same price.
The product matrix gave Dreame its first coverage of China’s complex channel system, but cross-channel product flows, distributor price cuts, and nationwide gray-market reselling directly created the next profit crisis.
29. Going from RMB300M to RMB3B still was not enough because China lost money
In 2022, China sales rose from roughly RMB300M to RMB3B and the category ranking reached number two or three, but the business still posted a small full-year loss. It dragged the company, which had previously made money through OEM work and overseas operations, into an overall loss.
郭人杰 was receiving offers to start a company or switch jobs, but he “didn’t think about leaving for even a second.” In his view of responsibility, scaling through cash burn and leaving immediately afterward meant the job was not finished.
This was the prehistory of “make money to fund the dream”: growth, brand, and market share cannot substitute for self-sustaining cash generation. Consumer hardware ultimately has to return to profit.
30. A 3-month profitability mandate slammed channels and expenses at the same time
俞浩 gave the China business 3 months to turn profitable. The team first suspended a large amount of B2B business, traced low-price inventory through machine serial numbers, and punished first-tier distributors whose controls failed rather than the street-level scalpers.
郭人杰 gives the example of a distributor that might earn RMB50,000 in a year but be fined RMB100,000 for price violations; some distributors left as a result. The cost was high, but market prices recovered visibly after roughly 1.5 months.
Channel fee rates were cut 30%–50%. The rule was: “Every activity must make money; if it doesn’t make money, don’t do it.” Outside observers began saying Dreame was going to shut down, so 郭人杰 and 俞浩 posted updates specifically to reassure partners.
He explains that income might fall 60% in the first month, 30% in the second, and turn positive in the third, because professional managers will find new efficiencies when judged on the annual result. “The potential of professional managers is infinite.”
31. Missing the profit deadline, 郭人杰 accepted a second demotion
Although the channel and expense reforms eventually worked, China was still not profitable at the 3-month deadline. 郭人杰 was therefore moved from horizontal leadership down to the robot-vacuum BU. He did not use the reform lag to exempt himself from the test.
His position was that if the company allowed him to eliminate teams based solely on results, it should evaluate him by the same standard. “I fully accepted it. I even believe in it.” Results discipline cannot apply only downward.
32. The robot vacuum with a robotic arm completed Dreame’s real premiumization
After moving to robot vacuums, 郭人杰 and R&D worked backward from a user pain point: traditional round bodies leave a cleaning blind spot of roughly 5 cm at right-angle edges because they must avoid collisions. The solution was to extend the mop outward on a small robotic arm.
The project was approved in March and launched in June, with roughly 3 months of development. The price rose from RMB3,599 to RMB5,299, and the product quickly proved out during 618.
郭人杰 says it lifted robot-vacuum share from a peak of roughly 5% to 22%. More importantly, it gave Dreame its first premium qualification based not merely on selling at a higher price, but on proprietary technology.
33. Facing low-price copies, Dreame proactively pushed the previous generation downmarket
Stone later put a similar function into a mid-priced product roughly RMB1,300 cheaper. Dreame’s response was not to protect the old flagship, but to upgrade the high end every 1 or 2 months while proactively moving the previous generation’s technology downmarket.
By that year’s Singles’ Day, robot-vacuum share was roughly 20% and floor-washer share also roughly 20%; both categories ranked second, making Dreame number one in China’s intelligent-cleaning market overall. 郭人杰 consequently returned to a higher management position.
34. Illness and a 1,000-person organization showed him his management ceiling
In June of the year before he left, 郭人杰 took roughly 1 month off because of an immune-system problem. At the same time, the organization had grown to more than 1,000 people, perhaps even 1,200–1,500, and he might directly handle problems from 50–60 people in a single day.
He says that most of his time felt like being “the neighborhood committee auntie”: from sitting down until midnight, every minute was automatically filled by meetings, emotions, and urgent issues. P&G taught him to understand mature systems, but not how to build a three-level organization from zero.
Looking back, he does not think Dreame had failed; he thinks his own capabilities were insufficient. Results-driven management could support roughly 500–600 people, but beyond that required a real organizational system.
35. “Make it number one” was both an ambition and a way to compress management complexity
郭人杰 believes asking sales teams to come in second leads to endless debates over RMB800M, RMB900M, and RMB910M. Asking them to be number one requires only beating a competitor’s RMB1B by a little, prompting the team to answer directly what resources are still missing.
The number-one goal also strips relationships and factions of explanatory power. Teams can band together and still be number one; if they band together but fail to deliver, they leave together. The rule helped him, while his management skills were still immature, survive rapid expansion.
The cost was equally clear: extremely hard targets, extremely rapid attrition, and an organization that expanded and contracted sharply. He sees Dreame’s high turnover as cultural selection—a culture not designed for people who treat work as simply “going to a job.”
36. Lexiang began as a Business Model startup, not a technology invention
Before leaving Dreame, 郭人杰 had planned to join IDG as an ER, or entrepreneur-in-residence, and search for a direction there because he did not want to be distracted while still at Dreame. Only after formally leaving did he systematically scan AI and robotics opportunities.
He defines his strengths as 3 things: high-frequency iteration; sensitivity to new sources of arbitrage; and experience defining, engineering, and selling consumer products from one to ten. Accordingly, he explicitly says he will not pursue “absolute innovation,” because “I don’t have 0-to-1 ability.”
His criteria were early PMF with substantial room to extend from “0.5 to 10”; global sales potential, allowing overseas traffic to absorb domestic R&D; and a combination with AI and robotics that could extend the short, 2-month window before traditional hardware is copied.
37. AI toys were only the entry point; the deeper observation was simultaneous improvement in play and retention
After scanning projects from the previous 3 years, 郭人杰 saw an opportunity in robot toys. Small robots could solve the limited play patterns of traditional toys, while AI could ease the low retention problem of being “thrown away after 2 plays.”
For roughly the first 2.5 months, the team worked on a small robot that could move freely across many environments. It needed high degrees of freedom and a highly malleable exterior; large sheet-metal surfaces could support clothing and graffiti, leaving room to layer on IP and new play patterns later.
He soon found that “making an AI toy” was not an accurate label. Robots and AI were not entering a unified toy category; they were redefining why people were willing to spend time.
38. LEGO confirmed that “toy” is an adjective, not a category
郭人杰 placed trading cards, character merchandise, mini 4WD cars, blind boxes, building blocks, and dolls side by side and concluded: “Toy is not a category. It’s just an adjective.” It describes something people are willing to spend time on for emotional value.
He sees 2 distinctive needs in LEGO: parents recognize the creativity, and finished creations can be displayed, taken outside, or posted on social media. It creates “creativity that parents can endorse and share.”
程曼祺 compared Minecraft to a virtual LEGO. 郭人杰 accepted the creativity analogy, but noted that sharing still requires screen recording or video editing. If a robot made its movement, personality, and appearance creatable, the barrier to sharing might be lower than for virtual work.
39. Reinforcement learning lets ordinary users define robot movement
郭人杰 compares reinforcement learning with learning to ride a bicycle. People do not first write out a complete movement sequence; they fall 10 times, then stop falling on the 11th. With a reward function, learning could eventually be more scalable than line-by-line programming.
If movement can be trained, users could have the robot throw a ball to a cat all day. Cameras could record the interaction and automatically generate a video. With AI, emotions could also be assigned: obedience, refusal, or even “talking back” could become user choices.
His end-state is “definable movement, definable emotion, and definable appearance,” combined with outdoor mobility and video sharing, turning the robot into a stronger creativity platform than static building blocks.
40. A hackathon with K-Scale Labs unexpectedly uncovered a low-cost sim-to-real market
Lexiang handles the body, hardware, and foundational algorithms, while Silicon Valley team K-Scale Labs handles reinforcement learning. 郭人杰 says K-Scale Labs had YC backing and originally targeted larger robots; the 2 sides immediately clicked over the small robot as a training platform.
He carried 2 robots to Silicon Valley for debugging, then shipped another 15 to 15 hackathon teams for free experimentation. Participants included PhDs, high-school students, and university students.
At a Berkeley lab, he saw Unitree robot dogs in widespread use but heard 3 pain points: roughly $30,000 was too expensive, buyers commonly had to wait 3 months, and cross-border repairs after a crash were slow. Researchers therefore hesitated to try complex movements.
41. The $1,000 developer robot could keep the company alive, but would not be the endgame
Lexiang found that a servo-driven small robot could not do front or back flips, but could walk, perform push-ups, and recover from a fall. At roughly $1,000 it could still be profitable while making sim-to-real affordable to students for the first time.
At the hackathon, one participant 3D-printed 2 hands and used Vision Pro for remote grasping control. Another trained the camera on human movement, allowing children to interact naturally and even teach the robot the Eight Brocades exercise. The team later absorbed 1 or 2 of these functions.
The standalone site already had roughly 30,000 lead-capture emails, with another 2,000–3,000 users entering communities. The product was expected to launch in the third quarter of the year referenced on the show, and 郭人杰 estimated that monthly sales of 2,000–3,000 units could support the company’s operations.
His genuine disappointment was: “We made such a great technological breakthrough, and in the end it seems to be only an educational tool.” That realization forced him to abstract from a single product toward a larger consumer scenario.
42. “Mobility plus anthropomorphism” became the 2 foundational capabilities of a home terminal
郭人杰 believes mobile bases still face 3-cm and 5-cm thresholds: balconies, bathrooms, low spaces, and stairs can all block whole-home mobility. The first value of human- or animal-like structures is their ability to cross spaces.
Anthropomorphism pulls functionality out of cold hardware. The same ancient poem played on a 12-inch screen feels different when a Nezha character tells it to a child with a voice and a face; AI can go further by generating custom voices and assigning emotion and personality.
Lexiang therefore defines the endgame as a home “mobile hardware terminal” that could carry information like a phone and support companionship, education, and pet interaction like an app platform.
43. Information exchange with elderly people is the most concrete first hypothesis for the home terminal
When 郭人杰 discussed financing with his mother, her first reaction was, “Are you going to get arrested?” She did not care about the amount. He saw this as an information-context gap among family members: children rarely have the patience to explain financing, equity, and actual risk from the beginning.
An AI device that can gradually add context could explain the information over several days, allowing the mother and son to have a conversation from a similar information base. The robot could also follow an older person and display the caller on its screen while maintaining communication when the user is sitting or lying down.
After the call, AI could continue explaining the weather and likely sights in Guangdong based on the child’s mention of a trip there. 郭人杰 does not deny that a phone can do this; he is betting that mobility, following, and anthropomorphism will increase older people’s willingness to interact.
44. The home robot will not promise chores yet; safety and experience are the initial hardware bar
程曼祺 noticed that 郭人杰 had not mentioned sweeping floors or folding clothes. He acknowledged that those tasks require very high technical capability, and said the current definition is closer to information exchange and companionship than a general-purpose household worker.
Even for a phone call, the platform must keep up with an elderly user, move quietly enough to reduce footstep and gear noise, and maintain the right eye-level or upward angle. It must also deliver the experience at a limited price rather than cost “100 times more” than a phone.
郭人杰 admits that the home scenario is “too new; nobody in history has done exactly the same thing.” The experience and demand still need to be validated through the platform and subsequent exploration.
Lexiang might run its first home exploration in November–December of the year referenced on the show, at a scale ranging from 50 units to 500 or even 500,000. The test is whether the platform is good enough first; only then can failure establish that the demand was wrong rather than the robot.
45. The WALL-E tracked robot extends the indoor terminal into yards and outdoors
Lexiang’s other product uses tracks to improve stability across outdoor terrain and recreates WALL-E from the film. 郭人杰 says the team built a prototype and signed the WALL-E IP; Disney considered it one of the better recreations they had seen. Launch was planned for September–October of the year referenced on the show.
The first hypothesis is home outdoor security. Mobile cameras might cover positions fixed surveillance cannot see and, unlike a guard dog, would not need to rest, but the actual performance remains to be validated.
The second hypothesis is child following and safety. A cartoon character may be easier for children to accept than a conventional robot dog and could accompany them to and from school or during outdoor activities. Display and entertainment are also possible purchase motives.
46. Running multiple products in parallel is an active testing strategy, but increases the R&D constraint
程曼祺 cited Li Zexiang’s advice that student entrepreneurs should ideally break through one category first, then asked why Lexiang was planning developer humanoid platforms, home terminals, and tracked robots at the same time.
郭人杰’s answer was that Lexiang is not a student-style single-technology startup, but Business Model-driven. The team will control R&D investment in each product and validate quickly through user insight and engineering, rather than defaulting to years of polishing one product.
This extends Dreame’s rapid-testing logic, while making R&D spending and organizational execution constraints under parallel development. 郭人杰’s defense remains that the company “admits mistakes quickly,” but Lexiang has not yet proved the model repeatable through public sales results.
47. A 39-person team is trying to combine robot-vacuum engineering with new robotics capabilities
At the time of recording, the company had roughly 39 people, including 32–33 in R&D. About 12–13 came from the robot-vacuum industry and brought mature capabilities in embedded systems, quality, and engineering; many of the rest came from large companies or were newly graduated master’s and PhD students.
The CTO had spent roughly 22 years at Ecovacs and had served as CTO. What 郭人杰 values most is not experience alone, but openness to new forms, because “experience” is naturally in tension with testing all 10 paths.
The gaps are concentrated in small-robot motion control, vision, mechanical design, and emotional models. 郭人杰 recruits from large companies such as ByteDance people who want to turn research into products, while also trusting young graduates with new modules.
48. On partnerships and talent, he emphasizes honest gap-filling with uncompromising results discipline
程曼祺 raised a common investor concern: could placing a core capability such as reinforcement learning at K-Scale Labs create long-term dependence? 郭人杰’s answer was, “If you can’t do it, you can’t do it.” Solving the problem quickly through partnership is more honest than pretending to be full-stack.
He expects both sides to fill their software and hardware gaps over time. Lexiang is building a control-algorithm team, while the US team will add hardware capabilities. For now, K-Scale Labs has photos of Lexiang’s engineers on its office wall and says its structural-innovation speed may be 3x what they imagined; the 2 sides serve as benchmarks for each other.
Lexiang’s values are “integrity, excellence, and openness.” 郭人杰 rejects equating elite schools with results and quotes, “Are kings and nobles born that way?” But for the 2 most specialized roles—motion control and emotional models—he still insists on finding the strongest candidates in terms of origin, background, and ability.
Non-specialist roles may get only 1 month to prove themselves. If they cannot deliver, they leave. He stresses that résumé-blind hiring must be paired with an absolute results-based elimination mechanism; otherwise the organization merely retains the hiring risk.
49. The faster the category converges, the closer it gets to the competitive phase 郭人杰 knows best
Facing consumer-robotics startups such as Vita Dynamics, Light Robotics, and Miaodong, as well as possible entry by large robotics companies, 郭人杰 welcomes more competitors. The more companies enter, the more they validate the direction; once product forms begin to converge, he enters the familiar phase of rapid iteration and direct competition.
His prerequisite remains “survive to the end.” Large robotics companies may not care about a market of 2,000–3,000 units a month, but that scale could first keep Lexiang alive. The developer product keeps the company alive; the home terminal carries the larger dream. The order cannot be reversed.
郭人杰 divides Chinese hardware into 3 generations: first, unable to make foreign equivalents; second, making equivalents cheaply and selling overseas; third, building better global products through Huawei, DJI, Dreame, Stone, and others. The next task is like Bambu Lab reshaping 3D printing: “Define what good means.”
His goal for the year was not to announce an ambitious sales target, but to “define a generation of product that we ourselves are satisfied with.” As long as the team gets its first real user feedback, he believes it can keep iterating. The long-term vision is “the greatest hardware company in the world,” but the operating core remains finding real pain points one by one and confirming that the products genuinely help people.